A Hybrid Spatiotemporal Deep Learning Model for Short-Term Metro Passenger Flow Prediction

Table 2

Parameter settings in the model construction.

Parameter	Description	Uniform random search interval	Optimal value
Convolutional filter
L	The length of convolution filter	(10, 40)	22
(a × b)	The size of convolution filter	—	(2 × 2)
Recurrent component
T	The number of predicted time steps in recurrent block	(5, 20)	6
H	The number of hidden units in the cells	(64, 128)	128
Optimizer
O	The selected optimizer during model training	(Adam, Nadam, RMSprop, and SGD)	RMSprop
α	The learning rate	(0.001, 0.01)	0.01
Training setting
D	The dropout rate	(0.2, 0.4)	0.2
B	The batch size for each training epoch	(20, 80)	35
E	The number of training epochs	(50, 200)	180